Only pass --global-base to the linker if its set to a non-default value #18056
Conversation
Otherwise rely on the linker default value of 1024. This change is in preparation of using `--stack-first` in debug builds. When `--stack-first` is passed to the linker `--global-base` is ignored (it should really be an error do this). I also avoid needless passing redundant arguments to the linker.
sbc100
force-pushed
the
default_global_base
branch
from
485de9c to
15255eb
Compare
| @@ -225,7 +225,7 @@ def lld_flags_for_executable(external_symbols): | |||
| cmd.append('--max-memory=%d' % settings.INITIAL_MEMORY) | |||
| elif settings.MAXIMUM_MEMORY != -1: | |||
| cmd.append('--max-memory=%d' % settings.MAXIMUM_MEMORY) | |||
| if not settings.RELOCATABLE: | |||
| if settings.GLOBAL_BASE != 1024 and not settings.RELOCATABLE: | |||
There was a problem hiding this comment.
Is it worth baking in current defaults of wasm-ld here? I worry about them changing. It seems safer to me to keep passing this.
There was a problem hiding this comment.
Yes, you are probably right there.
I guess part of my intention here is to express that case when the user doesn't care about the global base at all.. which is the normal case.
On the other hand, since you described the low memory optimization that binaryen does I think emscripten does actually care in more cases than I had thought.
I guess what I want here is to say "If the user specified a specific global base, or if emscripten want to run the low memory optimization, then pass an explictic stack base... otherwise we don't care and wasm-ld can choose whatever it wants"
There was a problem hiding this comment.
I see. Yeah, that makes sense. How about having a place that decides whether we use the low-memory optimization, and it sets GLOBAL_BASE if so? (to 1024) And if not it would be null. Then we could avoid passing null to wasm-ld.
Otherwise rely on the linker default value of 1024.
This change is in preparation of using
--stack-firstin debug builds. When--stack-firstis passed to the linker--global-baseis ignored (it should really be an error do this).I also avoid needless passing redundant arguments to the linker.